Orbital ordering in the layered perovskite material CsVF4

Ling-Fang Lin,1 Nitin Kaushal,1, 2 Yang Zhang,1 Adriana Moreo,1, 2 and Elbio Dagotto1, 2 1Department of Physics and Astronomy, University of Tennessee, Knoxville, Tennessee 37996, USA 2Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA In strongly correlated electronic systems, several novel physical properties are induced by the orbital degree of freedom. In particular, orbital degeneracy near the Fermi level leads to spontaneous symmetry breaking, such as the nematic state in FeSe and the orbital ordering in several perovskite 2 systems. Here, the novel layered perovskite material CsVF4, with a 3d electronic configuration, was systematically studied using density functional theory and a multiorbital Hubbard model within the Hatree-Fock approximation. Our results show that CsVF4 should be magnetic, with a G-type antiferromagnetic arrangement in the ab plane and weak antiferromagnetic exchange along the c- axis, in agreement with experimental results. Driven by the Jahn-Teller distortion in the VF6 1 1 octahedra that shorten the c-axis, the system displays an interesting electron occupancy dxy(dxzdyz) corresponding to the lower nondegenerate dxy orbital being half-filled and the other two degenerate dyz and dxz orbitals sharing one electron per site. We show that this degeneracy is broken and a novel dyz/dxz staggered orbital pattern is here predicted by both the first-principles and Hubbard model calculations. This orbital ordering is driven by the electronic instability associated with degeneracy removal to lower the energy.

I. INTRODUCTION crystal-field splitting is reversed as compared with expec- tations from an elongated c-axis. The resulting orbital Perovskites have attracted considerable interest for arrangement is a prerequisite for possible orbital order- decades because of their complex physical properties and ing in α-Sr2CrO4 [26]. extensive application values. In these strongly correlated Considering the physical and structural similarities systems, several physical degrees of freedom, such as spin, with the RP layered perovskites, analogous orbital or- charge, lattice, and orbital, are simultaneously active, ei- dering should also be obtained in DJ layered perovskites. ther cooperating or competing. This induces exotic phys- However, to our best knowledge, there are no orbital or- ical effects, such as colossal magnetoresistance in man- dering experimental results reported in DJ systems. For ganites [1, 2], magnetoelectricity [3, 4], electronic phase this reason, finding orbital ordering physics in the DJ separation [1, 5–8], and orbital ordering [9–12]. family from the theoretical perspective could play an im- Among peroskites, layered perovskite compounds are portant role in unifying the physical mechanisms between remarkable because they retain the essential features of DJ and RP layered perovskites. the perovskite structure while offering higher tunability From the known electronic occupation configuration and new capabilities induced by their low-dimensional of α-Sr2CrO4, it is reasonable to assume that finding the properties. The layered perovskite materials are formed same 3d2 electronic occupation in a DJ system defines by slicing perovskite slabs and inserting additional a feasible path to obtain DJ orbital-ordering physics. species in between layers. Among the known families of Hence, the layered perovskite compound CsVF4 with V layered perovskites, there are the two major structural 3d2 configuration, the simplest DJ family member cor- categories: the Ruddlesden-Popper (RP) [13, 14] and the responding to n = 1, captured our attention [28]. For- Dion-Jacobson (DJ) [15, 16] families, with general formu- tunately, for CsVF4 there is considerable and important 0 0 experimental progress. First, successive structural phase las A2An−1BnX3n+1 and A An−1BnX3n+1, respectively. These layered perovskite systems indeed have many in- transitions have been reported for CsVF4 in many inves- teresting physical properties. For the simplest n = 1 RP tigations and the corresponding details from high to low 1 case, unconventional high temperature superconductivity temperature are as follows: D4h (phase I, space group 0 0 0 7 was discovered in doped La2CuO4 [17] and in Sr2RuO4 [SG]: P4/mmm, a a c ) D2h (phase II, SG: Pman, − − 0 7 → + + 0 [18]. The so-called hybrid improper ferroelectricity was ap ap c ) D4h (phase III, SG: P4/nmm, ap ap c ) 13 → + + + 3 → arXiv:2011.10842v2 [cond-mat.str-el] 1 Feb 2021 initial predicted in n = 3 RP layered perovskites [19] D2h (phase IV, SG: Pmmn, ap ap c ) D2 (phase V, + + + → and later confirmed experimentally [20–22]. In addition, SG: P21212, ap bp c ) [29, 30]. Second, a magnetic phase hybrid improper ferroelectricity, as well as spin helix ar- transition in CsVF4 occurs at about 43 K. Due to the rangements, were also proposed in the DJ family [23–25]. weak interplane coupling between the VF4 layers, the Recently, the simplest n=1 RP family member α- magnetic structures can easily change between G- and Sr2CrO4 was reported to display an interesting orbital C-antiferromagnetic (AFM) order, or a mixed state can ordering transition both on the experimental and the- be easily reached by applying magnetic fields [31, 32]. oretical sides, even though the precise orbital ordering The most important aspect to remark is that the 3d2 configuration is still under debate [26, 27]. In this mate- electronic configuration provides the natural conditions 4+ 2 rial, Cr has a 3d electron configuration and the CrO6 for orbital ordering considering the progress reached in octahedra is elongated along the c-axis. Surprisingly, the the study of α-Sr2CrO4. Thus, it is interesting to inves- 2

dxy orbital. The G-type antiferromagnetism is found (a) to be the magnetic ground state, with a very weak ex- change coupling interaction along the c-axis, consistent Cs with the expected layered structure. More interestingly, a novel staggered dyz/dxz orbital-ordering pattern is here predicted, both by DFT and by the model calculations, V originating in an electronic instability for the special oc- 1 cupancy state (dxzdyz) . In addition, this interesting orbital-ordering pattern is sensitive to the crystal struc- c F ture symmetry and could be finely adjusted by subtle dis- tortions of the VF bonds in the ab plane. Our prediction b a of orbital order in CsVF4 also establish similarities with the orbital order discussed in manganites and ruthenates [33–35]. (b) Bond Length (Å) 1.964

II. METHODS 1.969

In the DFT portion of the project, first-principles b 1.865 calculations were performed using the revised Perdew- Burke-Ernzerhof exchange-correlation density functional (PBEsol), as implemented in the Vienna ab initio Sim- c a ulation Package (VASP) code [36–38]. The total en- ergy convergence criterion was set to be 10−5 eV dur- (c) 3+ 2 V : d ing the self-consistent calculation and the cutoff energy used for the plane-wave basis set is 550 eV. Most cal- d3z2-r2 eg d culations were carried out with the experimental crystal Jahn Teller x2-y2 structure fixed, i.e. without atomistic relaxation, and Octahedral distortion the corresponding k-mesh employed was 4 4 3. When crystal field (c-axis) the relaxing procedure is turned on, all lattice× × parame- d /d ters and atomic positions were optimized to obtain the t yz xz 2g d ground state structures until forces became lower than xy 0.01 eV/A.˚ Both non-magnetic and the spin-polarized phases were considered in our calculation. To better FIG. 1. (a) Side and (b) top views of the atomic structures describe the electron correlation for the spin-polarized for CsVF4 at room temperature, respectively. Dashed rect- phase, the generalized gradient approximation plus the angles indicate the unit cells. Bond lengths are indicated. U (GGA+Ueff ) approach [39] was adopted. Following (c) Schematic diagram of the expected energy-level splitting, previous studies addressing orbital physics in YVO3 and according to the crystal-structure information available for LaVO3 [40], the effective Hubbard coupling was fixed to CsVF . 4 the value Ueff = U J = 3 eV for simplicity. Note that, − besides the correction parameter Ueff , the exchange inter- action is already accounted for within the spin-polarized tigate the DJ layered perovskite CsVF4, especially with GGA exchange-correlation potential component. Thus, regard to orbital ordering, from a theoretical perspective. it would be inappropriate to simply compare DFT results In this work, the electronic and magnetic properties, at some value of Ueff with special locations in the JH/U as well as orbital ordering, of the simplest n=1 DJ lay- and U/W parameter space, as used in the model part. ered perovskite CsVF4 will be studied theoretically by From the ab initio ground-state wave function, the using both density functional theory (DFT) and a mul- maximally localized Wannier functions [41] within the tiorbital Hubbard model within the Hatree-Fock approx- orbital basis dxz, dyz, and dxy for each V ion were con- imation. Due to the very weak interaction between structed using the WANNIER90 code [42]. Based on our planes in the layered structure of CsVF4, this com- well-converged ab initio calculation, the relevant hopping pound can be regarded as an ideal platform for quasi- amplitudes and crystal-field splitting energies were ex- two-dimensional lattice models. Our first-principles re- tracted for the active t2g orbitals. Then, the ground-state 3+ sults indicate that the t2g orbitals of V display two phase diagram was investigated using the Hartree-Fock one-dimensional bands originating from the dxz/dyz or- method based on the multi-orbital Hubbard model dis- bitals and one two-dimensional band dominated by the cussed in Sec. V. 3

III. LATTICE PROPERTIES ( a ) D F T ( b ) 3 D F T

According to experimental studies, there are at least d x 2 - y 2 d z 2 five phases of the CsVF4 compound, as mentioned 2 ) d x z d y z above [29, 30]. However, only the crystal structure in- V e

( d 1 x y

formation measured at room temperature (phase IV) E is available, corresponding to the orthorhombic symme- try (SG: Pmmn) with the lattice constants a = 7.767, 0 b = 7.766, and c = 6.574, in units of A˚ [30]. We have - 1 tried to construct the crystal structure with phase V (SG: Γ X S Y Γ - 5 0 5 1 0 1 5 2 0 2 5 3 0 D O S P21212) for CsVF4 based on the information for RbFeF4 at room temperature [43] because it is is isostructural ( c ) T B ( d ) T B - u n f o l d 0 . 5 to CsVF4. However, during the DFT calculation pro- cess, this initial phase V structure becomes unstable and )

eventually converges to the phase IV crystal structure. V

e 0 . 0 ( According to a previous study [44], high values of the Coulomb repulsion are important to stabilize the low- E temperature distorted structures. Hence, our parame- - 0 . 5 ter Ueff was increased up to the range 4-9 eV for fur- ther testing. However, the phase V structure is still un- Γ X S Y Γ Γ' X ' S ' Y ' Γ' stable during the optimization process even in this new k y Y ' π S ' range. Thus, almost all of our calculations were per- / b ' ( e ) ( f ) formed based on the fixed experimental structure (phase IV, SG: Pmmn) obtained at room temperature [30]. b As shown in Figs. 1 (a) and (b), the CsVF4 compound Γ' b ' 0 X ' exhibits a single perovskite [VF4]∞ layer of corner-shared t a ' + b VF6 octahedra separated along the c-axis by the Cs cation, forming an infinity sandwich-like structure. The b t a tilted [VF6] octahedra system is in a configuration clas- k a π x sified as a+a+c+ according to Glazer notation [45]. For - / b ' c a p p - π/ a ' 0 π/ a ' each layer, the unit cell includes 4 V sites due to the ro- tation of the [VF ] octahedra. In contrast to -Sr CrO , 6 α 2 4 FIG. 2. (a) Band structure and (b) projected DOS of CsVF4 apparently each [VF6] octahedron in CsVF4 is shorten from DFT calculations for non-magnetic metallic phase. (c) along the c axis, while there exists only a small discrep- Tight-binding band structure. (d) Tight-binding unfolded ancy between the a and b axes. band structure. (e) Two-dimensional FS at the kz = 0 plane According to this structural information, the sketch in the unfolded BZ. (f) Sketch of the normal unit cell and the of the expected energy-levels splitting is indicated in unfolded one. Hoppings are also indicated. Fig. 1 (c) (the small discrepancy between a and b axes is ignored). Starting from the ideal cubic structure, once the [VF6] octahedron is formed, the energy levels of the The band structure and projected density of states five d orbitals splits into doubly degenerate eg orbitals (DOS) from the DFT calculations corresponding to the and triply degenerate t2g orbitals. Because the [VF6] oc- V atom’s 3d orbitals of CsVF4 are shown in Figs. 2 (a- tahedron is shorten along the c-axis, the triply degenerate b). Clearly, the states near the Fermi level are mainly t2g levels further split into a low-energy non-degenerate contributed by the t2g orbitals of the V ions, while the eg dxy orbital and two higher-energy doubly-degenerate dxz orbitals are located at a higher energy with broader band- 3+ 2 and dyz orbitals. V in this material has a 3d elec- width. All the three t2g bands crossing the Fermi energy tronic configuration. Therefore, it is natural to expect indicates that CsVF4 is a prototypical multiband system. that one electron always occupies the lowest dxy orbital Specifically, the energy level of the non-degenerate dxy or- while the other one is shared among the doubly degener- bital is lower than that of the two degenerate dxz and dyz ate dxz and dyz orbitals, which implies that the orbital orbitals, consistent with previous analysis of energy-level degree of freedom becomes indeed active. splitting, as shown in Fig. 1 (c). As shown in Fig. 2 (c), the DFT bands are fitted very well by the tight-binding (TB) bands of the three molec- IV. FIRST-PRINCIPLES CALCULATIONS ular orbitals obtained from the maximally localized Wan- nier functions. From the above TB fitting, the crystal- Non-magnetic Metallic Phase. Let us start with the field levels of the t2g orbitals are ∆xz = 0.176 eV, ∆yz hypothetical non-magnetic metallic phase of CsVF4 ob- = 0.176 eV, and ∆xy = -0.004 eV while the associated tained under the assumption of no spin polarization. hopping amplitudes in the dxz, dyz, dxy orbital basis { } 4 are

 0.126 0 0  − t~a =  0 0 0  , J1 0 0 0.143 − (1) J 0 0 0  2

t~b = 0 0.126 0  . 0− 0 0.143 − c Here, only the nearest-neighbor hoppings and amplitudes of hoppings larger than 0.1 eV are considered for simplic- (a) FM (b) A-AFM a b ity. To capture the degenerate properties of the dxz and dyz orbitals, reasonable modifications of the hopping pa- rameters are adopted. Interestingly, there are four V in the primitive unit cell in the DFT calculation. Thus, we can further simplify the current band structure by unfolding the Brillouin zone (BZ) [Fig. 2 (f)]. The unfolded band structure and cor- responding two dimensional Fermi surface (FS) can be found in Figs. 2 (d-e). According to the band struc- ture, it is interesting that the dxy orbital band is broad along all high-symmetry paths in the xy-plane, displaying (c) C-AFM (d) G-AFM quasi-two-dimensional (2D) properties. By comparison, the dxz and dyz orbitals show almost flat bands along certain high-symmetry directions Γ0-Y0 (X0-S0) and Γ0- FIG. 3. The various magnetic configurations studied in this X0 (Y0-S0), respectively, exhibiting quasi-one-dimensional work. (a) FM, (b) A-AFM, (c) C-AFM, and (d) G-AFM. J1 and J2 are the ab-plane and c-axis exchange couplings, (1D) properties. Clearly, the 2D FS consists of two quasi- respectively. 1D bands and one quasi-2D band. The quasi-1D bands originate from the dxz and dyz orbitals, while the quasi- 2D band is dominated by the dxy orbital. This interesting TABLE I. List of energy equations and calculated energies of result is quite similar to that known to occur in the chiral the four collinear spin configurations used to determine the p-wave superconductor Sr2RuO4 [46]. Meanwhile, due to magnetic exchange integrals. The G-AFM state is taken as the 2D (1D) characteristic of the dxy (dxz/dyz) orbital, the reference of energy. its bandwidth is broader (4tdxy ) than that of the dxz/dyz orbitals (2t ) as shown in Figs. 2 (c-d). Confg. Energy equations Energy (meV/f.u.) dxz /dxz 2 2 Magnetism. In the following DFT calculations, now FM E0 − 2J1S − J2S 16 2 2 the spin polarization is allowed. Four magnetic struc- A-AFM E0 − 2J1S + J2S 16 2 2 tures [i.e., ferromagnetic (FM) and A-, C-, and G-type C-AFM E0 + 2J1S − J2S 1 2 2 AFM states, see Fig. 3] were calculated using the fixed G-AFM E0 + 2J1S + J2S 0 atomic experimental structure discussed before. The cor- responding energies are list in Table. I. From this infor- mation, the exchange interactions can be estimated by capture the main physics. According to our calculations mapping the calculated total energies for each magnetic below, the orbital ordering patterns are not sensitive to state to the Heisenberg model. The nearest-neighbor ex- having G- or C-AFM magnetic order. We also tested change coupling constants can be extracted using whether the FM spin order has an effect on the orbital 1 ordering. The results show that the orbital ordering is J1 = 2 [E(F ) E(G) E(C) + E(A)], independent from the spin order, see the appendix Fig. 9 −8S − − (2) 1 for more details. Thus, for simplicity, the C-AFM order J2 = 2 [E(F ) E(G) + E(C) E(A)], is considered in the following calculations, unless other- −4S − − wise stated. where S = 1 is the magnetic moment. The extracted re- Orbital ordering. The calculated results for the pro- sults are J1 = 3.9 meV and J2 = 0.3 meV indicating jected DOS are shown in Fig. 4, where we find that the that both the ab−-plane and c-axis favor− AFM couplings. states near the Fermi level mainly contribute from the As expected, due to the layered structure of CsVF4, J2 t2g orbitals of the V ions while the eg orbitals are located is very weak and can be neglected, agreeing well with at higher energies (not shown here). As expected, the experimental investigations [31, 32]. Since the interplane dxy orbitals are always occupied by one electron, while coupling is weak and we only focus on the intrinsic prop- the combination of degenerate dxz and dyz orbitals is oc- erties of each layer, in practice either G- or C-AFM can cupied by another electron. An electronic instability is 5

(e) V1 V4 V1 Bond Length (Å) 1.964 V3 V2 V3 1.969 b 1.865 V1 V4 V1 (e) (along c-axis) V2 V1 V2 c a y Bond Length (Å) FIG. 4. (a)-(d) Calculated partial DOS’s projected onto the b V1 V2 V1 five d orbitals of four V ions based on the fixed experimental 1.979 crystal structure with C-AFM. The vertical dash line in each c x panel represents the Fermi level. ↑ (↓) represents spin up 1.903 (down). (e) Charge density at the region extending from - a V2 V1 V2 (along c-axis) 0.2 eV to the Fermi level.

expected to occur when two orbitals share one electron. FIG. 5. (a)-(d) Calculated band and partial DOSs projected Therefore, linear combinations of dxz and dyz lead to onto the five d orbitals of four V ions based on the relaxed two separated states, the occupied and unoccupied lev- high-symmetry crystal structure. The horizontal dashed line els, opening a large band gap of about 2 eV. The physical in each panel represents the Fermi level. ↑ (↓) represents spin reason for the band gap splitting is that the formation up (down). (e) Charge density at the region extending from of the orbital ordering (OO) pattern breaks the symme- -0.1 eV to the Fermi level. The local axes x, y, and z are defined as the [110], [110],¯ and [001] directions of the unit try to lower the system’s energy, no matter what kind of cell. pattern it forms. Here the large gap (2 eV) is related to the strong electronic correlation in this material, namely, the parameter . A larger corresponds to a larger Ueff Ueff the VF6 octahedra in the ab plane. A larger cell size band gap, which is also in good agreement with the re- (√2 √2 1) is adopted here as compared to the min- sults of the Hartree-Fock model portion of this publica- imal× one (one× formula unit per cell, a = b = 3.958 and tion. The charge density for the occupied states are visu- c = 6.546 in units of A˚ ) so as to allow for symmetry- ally provided in Fig. 4(e), displaying a staggered orbital breaking distortions. To remove interference factors, all ordering. Even though there are small discrepancies be- lattice parameters and atomic coordinates are optimized tween the a- and b-axis lattice lengths, this anisotropy in for the HSS using DFT. The self-consistent calculated the electronic structure and orbital ordering along the a results are shown in Figs. 5 (a-d), with the nearly flat and axes can be neglected. From the symmetry point of b bands of dxz/dyz orbitals indicating strongly localized view, when only considering the crystal symmetry (phase electronic behavior. Clearly, electrons near the Fermi IV, SG: Pmmn), the V1, V2, V3, and V4 atoms are equiv- level occupy dxz/dyz orbitals in a staggered manner be- alent and the corresponding Wyckoff position is (0,0,0). tween all nearest-neighbor V atoms in the ab-plane, lead- To determine whether the electronic instability in- ing to staggered orbital ordering pointing toward orthog- duced orbital ordering is intrinsic or not, we construct a onal directions, as shown in Fig. 5 (e). In summary, the high symmetry structure (HSS) [phase I, SG: P4/mmm, spontaneous electronic instability unveiled here breaks 0 0 0 a a c )] to remove all the distortion and rotation of degeneracies of the dxz/dyz orbitals, resulting in stag- 6

(a) SG: Pmmn 0.30

Bond Length (Å) 0.28 PM/IC-SDW (M) FM (M)

1.990 0.26 FM+AFO (I)

/U 0.24 H

1.992 J 0.22

1.905 0.20 AFM+AFO (I) 0.18

0.16 (a) 0.3 0.5 1.0U/W 2.0 3.0 4.0 6.0

38 meV FIG. 7. Phase diagram of the three-orbital Hubbard model (b) varying the Hund JH /U and Hubbard U/W couplings, with W the bandwidth. Calculations were performed for all the (b) SG: P2/c points shown, using a cluster size 12 × 12. In this cluster, the bandwidth is W = 1.145 eV. The notation PM, IC-SDW, FM, Bond Length (Å) AFM, AFO, M, and I stands for paramagnetic, incommen- surate spin density wave, ferromagnetic, antiferromagnetic, 1.959/1.954 antiferro-orbital, metallic, and insulator, respectively.

2.028/2.023 b the calculations, 2 electrons per site are considered. The 1.905 model studied here includes the kinetic energy and inter- action energy terms = + . The tight-binding c a H Hk Hint kinetic component is FIG. 6. Sketch of the (a) ferro- and (b) staggered-orbital X ~α † X ordering calculated with the DFT optimized structure. Hk = tγγ0 (ciσγ ci+~ασγ0 + H.c.) + ∆γ niγσ, (3) iσ iγσ ~αγγ0 gered orbital ordering in CsVF4. where the first term represents the hopping of an elec- As mentioned before, there is a small discrepancy be- tron from orbital γ at site i to orbital γ0 at the nearest- tween the a- and b-axis if the experimental structure neighbor site i + ~α. The vector ~α connects nearest- would be used. If we relax all lattice parameters and neighbor sites along the ~a and ~b axes, namely ~α is the atomic coordinates, then the formation of ferro-orbital unit vector either along the x or y axis with length a ordering is realized by the reinforced anisotropy [Fig. 6 and b, respectively. γ and γ0 represent the three different (a)]. Similarly, starting from the fixed experimental orbitals d , d , d . ∆ is the crystal-field splitting of structure, if the VF bond length ab-plane is tuned by xz yz xy γ orbital γ. The actual values for the hopping matrix and hand with alternating nudged amplitudes, the symmetry crystal-field splittings are extracted from the ab initio of the relaxed structure is lowered (SG: P2/c) as exhib- calculations, as described in the previous section. ited in Fig. 6 (b). The staggered orbital ordering is rein- The electronic interaction portion of the Hamiltonian forced and lowers the total energy by 38 meV. According is: to this interesting observation, it is reasonable to specu- late that the orbital ordering patterns are very sensitive X 0 JH X to the crystal structure and controllable by fine tuning, Hint = U ni↑γ ni↓γ + (U ) niγ niγ0 − 2 such as via strain. iγ i γ<γ0 X X † 2JH Si,γ Si,γ0 + JH (P Piγ0 + H.c.). (4) − · iγ V. HUBBARD MODEL i i γ<γ0 γ<γ0

Due to the weak interaction between layers in the The first term is the standard intraorbital Hubbard repul- CsVF4 compound, for simplicity only the 2D square lat- sion. The second term is the electronic repulsion between tice for the ab plane will be considered in the electronic electrons at different orbitals where the standard relation 0 model. Specifically, an effective three-orbital Hubbard U = U 2JH is assumed. The third term represents the model for the two-dimensional square lattice will be con- Hund’s− coupling between electrons occupying the three structed to describe the spin and orbital orderings. In all active 3d orbitals. The operator Siγ is the total spin at 7 site i and orbital γ defined as 1.0 (a)JH /U = 0.2 2.0 1 X † 0.8 Siγ = c σσσ0 c 0 . (5) ) ) 2 iσγ iσ γ q q S(π, π) S 2 ( ( σσ0 0.6 τ(π, π) h i

,, τ τ 1.5

) ) 2

The fourth term is the pair hopping between different q q 0.4 S ( ( h i orbitals at the same site , where = . S S PM AFM+AFO i Piγ ci↓γ ci↑γ 0.2 The unrestricted real-space Hartree-Fock method is 1.0 applied to solve numerically the Hamiltonian we con- 0.0 1.0 structed [47]. We performed a Hartree-Fock decomposi- (b)JH /U = 0.25 2.0 tion on all the quartic fermionic terms in the interaction, 0.8 ) 2 q S(0, 0) S

leading to the single-particle density matrix elements ( † 0.6 τ(π, π) h i c ciσ0γ0 , as the mean-field parameters. Then, self , τ 1.5 iσγ ) 2 PM FM+AFOFM

h i q 0.4 S consistency in those mean-field parameters was achieved ( h i using the modified Broyden’s method [48]. The chemical S 0.2 1.0 potential µ was tuned to target the required electronic 0.0 density. Up to 15 random configurations of order pa- 0.1 0.5 1.0 2.0 5.0 rameters were used to start the iterative process to gain U/W convergence at every point, and the converged states with (c) (d) the lowest energy were chosen as the result. We calcu- AFM+AFO FM+AFO lated the local electronic density, density of states, local xy spin moment S2 , spin structure factor S(q), and orbital yz h i y xz structure factor τ(q) to identify the phases. We used the tyz following definitions: 1 X S2 = S2 , h i L L h i i x x y i txz 1 X iq·(ri−rj ) S(q) = 2 Si Sj e , FIG. 8. Panels (a,b) show the spin structure factor S(q), (LxLy) h · i i,j orbital structure factor τ(q), and averaged local spin mo- 1 ment hS2i, at J /U = 0.2 and 0.25. Panels (c) and (d) show X iq·(ri−rj ) H τ(q) = 2 τiτj e , (6) the pictorial represention of the AFM+AFO and FM+AFO (LxLy) h i i,j states, respectively. where τi = (ni,xz ni,yz)/2. Figure 7 shows− the full phase diagram of the three- orbital Hubbard model varying JH /U from 0.15 to 0.30 slightly greater than 2.0, because of considerable con- and U/W from 0 to 6. For small U/W 0.6, the sys- tributions arising from the states with three electrons at tem mostly shows paramagnetism and the≤ presence of the same site. incommensurate spin density wave order near the phase According to the full phase diagram, the AFM+AFO boundaries. Interestingly, for JH 0.24, and for most of state is predicted to be the most relevant phase for the the region of the phase diagram, we≤ found large peaks at real material, because the majority of the phase diagram momentum q = (π, π) in S(q) and τ(q) suggesting an- comprises of this state. Moreover, this is in good agree- tiferromagnetic spin ordering accompanied by antiferro- ment with the DFT results. The single-particle den- orbital ordering (namely a combined state AFM+AFO). sity of states is also calculated and the system is found In Fig. 8(a), we fix JH /U = 0.2 and show the evolution to be insulating in the AFM+AFO state. The average of S(π, π), τ(π, π), and S2 with U/W , to illustrate that orbital-resolved local density calculations show that the h i starting from intermediate Hubbard repulsion the system xy orbital is exactly half-filled (i.e nxy = 1), whereas h i is in an AFM+AFO state with robust local spin moments nxz = nyz = 0.5. In Fig. 8(c), a pictorial representa- and as we increase U/W further, S2 saturates to 2.0, tionh i of theh AFM+AFOi state is displayed. The antiferro- corresponding to spin 1 as expected.h i magnetic spin order is driven by the half-filled xy orbital As shown in the phase diagram, for JH /U 0.24 the with the largest hopping amplitude. These xy spins be- system mainly presents ferromagnetic ordering≥ (FM). In ing parallel to the spins on orbitals xz/yz because of Fig. 8(b), the evolution with U/W of S(0, 0), τ(π, π), the robust Hund’s coupling, create the spin 1 local mo- and S2 are shown. For intermediate values of U/W , ment. The staggered orbital ordering among the xz/yz a FM-metallich i region was found, while for large U/W , a orbitals is energetically preferred to ease the movement FM-insulator accompanied with antiferro-orbital order- of the electrons (i.e. decrease in kinetic energy). Note ing (FM+AFO) region is present. Once again, in the that if the Hund’s coupling is increased beyond 0.24, the FM+AFO region the spin-moment squared is saturated FM state is stabilized because now the Hund’s term play to value 2.0, whereas in the FM-metal region S2 is the dominant role in the energy of the intermediate state h i 8 via hopping of electrons [see Fig. 8(d)], as in the double- exchange mechanism. 2 ( a ) V 1 ( b ) V 2 d x 2 - y 2 d z 2 S

O 1 d x z d y z D VI. DISCUSSION d x y 0

The electron-electron interaction and electron-phonon V 4 2 ( c ) V 3 ( d ) coupling are the two major possible mechanisms to cause S the orbital ordering discussed here. But which one is the O primary cause? Both in the DFT and model portions D 1 of the manuscript, we constructed the high symmetry 0 structure, removing all the distortions and rotations in - 1 0 1 2 - 1 0 1 2 the ab plane to analyze the role of the electron-phonon E n e r g y ( e V ) E n e r g y ( e V ) coupling in the system. Interestingly, the results show ( e ) V 1 V 4 V 1 that the staggered orbital ordering is still robust. In B o n d L e n g t h ( Å ) other words, the AFO pattern dominates even when the 1 . 9 6 4 electron-phonon coupling is not included in the model V 3 V 2 V 3 we studied. These results indicate that electron-electron b 1 . 9 6 9 interaction is the intrinsic driving mechanism in CsVF . 4 V 1 V 4 V 1 1 . 8 6 5 However, we cannot establish if spin or orbital are the c a a l o n g c - a x i s main drivers of the symmetries broken. Both are en- tangled. Only a calculation including finite temperature can find out which of the two critical temperatures, i.e. FIG. 9. (a)-(d) Calculated partial DOS’s projected onto the Torbital or Tspin, occurs first upon cooling. Then that five d orbitals of four V ions based on the fixed experimental would establish which one is the “driver” and which one crystal structure with FM order. The vertical dash line in the “passenger”. But this calculation is very difficult, each panel represents the Fermi level. ↑ (↓) represents spin particularly within DFT, and it is postponed to future up (down). (e) Charge density at the region extending from -0.4 eV to the Fermi level. work. If the electron-phonon coupling would be included, how does this coupling affect the orbital ordering? Does VIII. ACKNOWLEDGMENTS this new coupling establish clearly whether spin or orbital dominate? These questions deserve further work. Typi- cally, orbital and lattice work together to induce orbital This project was supported by the U.S. Department of order and probably with phonons included, the orbital Energy (DOE), Office of Science, Basic Energy Sciences would be the main driver over spin. (BES), Materials Science and Engineering Division.

VII. CONCLUSIONS IX. APPENDIX

In this work, first-principles DFT and Hubbard model As shown in Figs. 9 (a-d), if the magnetism is fixed calculations for CsVF4 were performed. Due to the lay- to be FM, the charge density from the dxy and dxz/dyz ered structure of CsVF4, the coupling between inter- orbitals is not too different from the C-AFM case. Even planes is very weak and can be neglected. For this rea- though the final charge density shown in Fig. 9 (e) is son, the CsVF4 compound provides an ideal platform to mixed with some dxy orbital at the region in [-0.4, 0] eV, study quasi two-dimensional lattice models. Our theoret- it is clear that the dxz/dyz orbitals are showing the same 3+ ical results indicate that the t2g obitals of V dominate pattern as for C-AFM. In other words, the orbital order- and display two quasi one-dimensional bands originat- ing appears independent from the spin order. This result ing from the dxz/dyz orbitals and one two-dimensional is also consistent with our model calculations, where both band dominated by the dxy orbital. Furthermore, the the AFM+AFO and FM+AFO phases are shown in the G-type antiferromagnetism is found to be the dominant phase diagram to be stable at different regions. magnetic ground state, in agreement with previous ex- To understand why a U/W robust is needed for the or- perimental results. More interestingly, a novel staggered bital ordering, we can start with the atomic limit. Large dyz/dxz orbital ordering pattern is here predicted, driven U and robust Hund’s coupling (for example JH/U = 0.2) by an electronic instability for the special electron occu- will prefer that electrons are present in different orbitals 1 pancy state (dxz/dyz) . In addition, this orbital ordering but with the same spin. The lower energy of the dxy or- pattern is sensitive to the crystal structure symmetry and bital induces one electron to be located in the dxy orbital. could be finely adjusted by subtle distortions of the VF Now, if we turn on the kinetic energy term (for simplicity bond in the ab plane. we are using only two sites here) the effective superex- 9

ping is restricted only to the x direction because the y- direction hopping of the xz orbital is zero. The picture described above shows that the hoppings for CsVF4 fur- ther stabilizes the AFO state for a large range of U, in addition to the fact that in the large U limit the AFO 10.0 yz (a) 8.0 xz

) xy ω

( 6.0 ρ 4.0 2.0 0.0 FIG. 10. Pictorial understanding of the AFM+AFO states, 0.0 2.0 4.0ω 6.0 8.0 as discussed in the text. 0.04 Int. (b) 0.02 K.E. AF O change between the half-filled -sites lead to antiferro- Total

xy E magnetic ordering, while electrons in the dxz/dyz orbitals just follow the same spin ordering because of the robust − 0.00

Hund’s coupling. Now assuming this antiferromagnetic FO state, one of the two states in Figs. 10(a,b) is possible. E 0.02 The cost of hopping as in the arrow is smaller for state − in (b) because there is no double occupancy (the change in Hund’s coupling energy is ignored for simplicity, as it 0.04 will be same in both cases). The above discussion ex- − 1U/W 10 plains why the AFO+AFM state is favored. A similar argument can be used for the 2d lattice with the actual FIG. 11. (a) DOS for a 16 × 16 cluster in the AFM+AFO hopping terms used in the model; we have indeed com- state with U/W = 4.0. (b) The energies of the FO and AFO pared the energies of the different Ansatz states using Ansatz states, both with AFM ordering. our Hartree-Fock code to understand why AFO state is preferred. Figure 11(a) shows the DOS for a 16 16 cluster in the AFM+AFO state with U/W = 4.0. We× found a gap of nearly 1.6 eV, and we checked that the gap increases as we increase U. Thus, clearly correlations effects are responsible for the physics we found. We have calculated the energies of the FO and AFO Ansatz states, both with AFM ordering, as shown in panel (b). Please notice that the total energy of the FO state is higher than the AFO state, and the main reason originates in the higher kinetic energy in the FO state. Why the AFO state has lower kinetic energy i.e. why electrons move relatively easier in AFO than in FO? Intu- FIG. 12. The pictorial understanding of the AFO state using itive understanding can be gained by the cartoon shown a 2×2 cluster. in Fig. 12. We can focus only on the xz/yz orbitals, because the xy orbital behaves similarly in both states. In the AFO state electrons can hop in both directions, exchange is larger than the FO exchange, as discussed whereas in the FO state the drawn-above electron hop- above in Fig. 10. 10

[1] E. Dagotto, T. Hotta, and A. Moreo, Phys. Rep. 344, 1 [23] N. A. Benedek, Inorg. Chem. 53, 3769 (2014). (2001). [24] B.-W. Li, M. Osada, T. C. Ozawa, and T. Sasaki, Chem. [2] E. Dagotto, Science 309, 257 (2005). Mater. 24, 3111 (2012). [3] I. A. Sergienko and E. Dagotto, Phys. Rev. B 73, 094434 [25] C. Autieri, P. Barone, J. S lawi´nska, and S. Picozzi, Phys. (2006). Rev. Mater. 3, 084416 (2019). [4] I. A. Sergienko, C. S¸en, and E. Dagotto, Phys. Rev. Lett. [26] T. Ishikawa, T. Toriyama, T. Konishi, H. Sakurai, and 97, 227204 (2006). Y. Ohta, J. Phys. Soc. Jpn. 86, 033701 (2017). [5] A. Moreo, S. Yunoki, and E. Dagotto, Science 283, 2034 [27] Z. Zhu, W. Hu, C. Occhialini, J. Li, J. Pelliciari, C. Nel- (1999). son, M. Norman, Q. Si, and R. Comin, arXiv preprint [6] S. Yunoki, J. Hu, A. L. Malvezzi, A. Moreo, N. Furukawa, arXiv:1906.04194 (2019). and E. Dagotto, Phys. Rev. Lett. 80, 845 (1998). [28] P. Hagenmuller, Inorganic Solid Fluorides: Chemistry and Physics [7] T. Miao, L. Deng, W. Yang, J. Ni, C. Zheng, (Elsevier, 2012). J. Etheridge, S. Wang, H. Liu, H. Lin, Y. Yu, Q. Shi, [29] M. Hidaka, S. Yamashita, K. Inoue, N. Tsukuda, G. B. J., P. Cai, Y. Zhu, T. Yang, X. Zhang, X. Gao, C. Xi, and W. B. M., J. Phys. Soc. Jpn. 50, 4022 (1981). M. Tian, X. Wu, H. Xiang, E. Dagotto, L. Yin, and [30] M. Hidaka, H. Fujii, B. Garrard, and B. Wanklyn, Phys. J. Shen, Proc. Natl. Acad. Sci. USA 117, 7090 (2020). Status Solidi A 96, 415 (1986). [8] H. Lin, H. Liu, L. Lin, S. Dong, H. Chen, Y. Bai, T. Miao, [31] M. Hidaka, H. Fujii, M. Nishi, and B. Wanklyn, Phys. Y. Yu, W. Yu, J. Tang, Y. Zhu, Y. Kou, J. Niu, Z. Cheng, Status Solidi A 117, 563 (1990). J. Xiao, W. Wang, E. Dagotto, L. Yin, and J. Shen, [32] M. Hidaka, T. Ogata, T. Eguchi, H. Akiyama, and Phys. Rev. Lett. 120, 267202 (2018). B. Wanklyn, Phys. Status Solidi A 155, 519 (1996). [9] J. Varignon, M. Bibes, and A. Zunger, Nat. Commun. [33] T. Hotta, S. Yunoki, M. Mayr, and E. Dagotto, Phys. 10, 1 (2019). Rev. B 60, R15009 (1999). [10] H. Chen, D. P. Kumah, A. S. Disa, F. J. Walker, C. H. [34] T. Hotta and E. Dagotto, Phys. Rev. Lett. 88, 017201 Ahn, and S. Ismail-Beigi, Phys. Rev. Lett. 110, 186402 (2001). (2013). [35] C. S¸en and E. R. Dagotto, Phys. Rev. B 102, 035126 [11] J. Varignon, M. N. Grisolia, D. Preziosi, P. Ghosez, and (2020). M. Bibes, Phys. Rev. B 96, 235106 (2017). [36] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999). [12] H. D. Zhou, B. S. Conner, L. Balicas, and C. R. Wiebe, [37] P. E. Bl¨ochl, Phys. Rev. B 50, 17953 (1994). Phys. Rev. Lett. 99, 136403 (2007). [38] J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, [13] S. Ruddlesden and P. Popper, Acta Cryst. 10, 538 (1957). G. E. Scuseria, L. A. Constantin, X. Zhou, and K. Burke, [14] S. Ruddlesden and P. Popper, Acta Cryst. 11, 54 (1958). Phys. Rev. Lett. 100, 136406 (2008). [15] M. Dion, M. Ganne, and M. Tournoux, Mater. Res. Bull. [39] S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. 16, 1429 (1981). Humphreys, and A. P. Sutton, Phys. Rev. B 57, 1505 [16] A. Jacobson, J. W. Johnson, and J. Lewandowski, Inorg. (1998). Chem. 24, 3727 (1985). [40] Z. Fang and N. Nagaosa, Phys. Rev. Lett. 93, 176404 [17] P. W. Anderson, Science 235, 1196 (1987). (2004). [18] K. Ishida, H. Mukuda, Y. Kitaoka, K. Asayama, Z. Mao, [41] N. Marzari and D. Vanderbilt, Phys. Rev. B 56, 12847 Y. Mori, and Y. Maeno, Nature (London) 396, 658 (1997). (1998). [42] A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Van- [19] N. A. Benedek and C. J. Fennie, Phys. Rev. Lett. 106, derbilt, and N. Marzari, Comput. Phys. Commun. 178, 107204 (2011). 685 (2008). [20] Y. S. Oh, X. Luo, F.-T. Huang, Y. Wang, and S.-W. [43] M. Hidaka, H. Akiyama, and B. Wanklyn, Phys. Status Cheong, Nat. Mater. 14, 407 (2015). Solidi A 97, 387 (1986). [21] S. Yoshida, K. Fujita, H. Akamatsu, O. Hernandez, [44] C. Autieri, E. Koch, and E. Pavarini, Phys. Rev. B 89, A. Sen Gupta, F. G. Brown, H. Padmanabhan, A. S. 155109 (2014). Gibbs, T. Kuge, R. Tsuji, S. Murai, J. M. Rondinelli, [45] A. M. Glazer, Acta Cryst. B 28, 3384 (1972). V. Gopalan, and K. Tanaka, Adv. Funct. Mater. 28, [46] C. Kallin, Rep. Prog. Phys. 75, 042501 (2012). 1801856 (2018). [47] Q. Luo, A. Nicholson, J. Rinc´on, S. Liang, J. Riera, [22] M. Liu, Y. Zhang, L.-F. Lin, L. Lin, S. Yang, X. Li, G. Alvarez, L. Wang, W. Ku, G. D. Samolyuk, A. Moreo, Y. Wang, S. Li, Z. Yan, X. Wang, X.-G. Li, S. Dong, and E. Dagotto, Phys. Rev. B 87, 024404 (2013). and J.-M. Liu, Appl. Phys. Lett. 113, 022902 (2018). [48] D. D. Johnson, Phys. Rev. B 38, 12807 (1988).